On CI for sharded runs several tests causing complete. Agreed to make bug from these. Especially lately tests: igt@gem_fence_thrash@bo-write-verify-threaded-y igt@gem_mmap_gtt@forked-medium-copy igt@gem_mmap_gtt@forked-medium-copy-XY igt@gem_mmap_gtt@forked-medium-copy-odd igt@gem_mmap_gtt@medium-copy To get these sorted out we could see maybe more stable APL. See: https://intel-gfx-ci.01.org/tree/drm-tip/shards.html Note APL's currently without displays.
The main culprit there is the continual runtime suspend/resume making that platform much slower than the others; rpm is especially harsh towards GTT mmapings.
https://patchwork.freedesktop.org/series/29764/ to speed up mmap-gtt and https://patchwork.freedesktop.org/series/29208/ to speed up gem_fence_thash.
Patches sent here seems to fix issues seen on APL shards?
Adding more data links since we now track bugs for APL-shards: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3190/shard-apl8/igt@gem_fence_thrash@bo-write-verify-threaded-y.html https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3190/shard-apl6/igt@gem_mmap_gtt@forked-medium-copy.html https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3190/shard-apl1/igt@gem_mmap_gtt@forked-medium-copy-XY.html https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3190/shard-apl2/igt@gem_mmap_gtt@forked-medium-copy-odd.html
The situation has improved slightly by the runtime pm issues being masked by the connected display, and kernel commit f34a93bbb33d988df36bb99f791f0bf8a7017041 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Mon Oct 9 09:43:58 2017 +0100 drm/i915: Check PIN_NONFAULT overlaps in evict_for_node ... kernel commit 43ae70d97c5195f48d903df31ecac4c5397b2f1e Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Mon Oct 9 09:44:01 2017 +0100 drm/i915: Early rejection of mappable GGTT pin attempts for large bo will improve gem_mmap_gtt / gem_pread / gem_pwrite. gem_fence_thrash is improved by igt commit eb57c0289dbd2bfb573bb1c104d80eaf5edc7228 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Wed Aug 23 13:44:57 2017 +0100 igt/gem_fence_thrash: Use streaming reads for verify That should hopefully account for the remaining incompletes. Although as stated earlier, part of the problem is being masked by a connected display and we don't have good test coverage of GTT vs rpm (or at least that demonstrates the severe performance issues).
I see all those now passing with Fence: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3196/shard-apl8/igt@gem_fence_thrash@bo-write-verify-threaded-y.html And mmaps: on build: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3197/ https://intel-gfx-ci.01.org/tree/drm-tip/shards.html
Closing old verified.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.